The world's largest music publishers have written to Google, Microsoft and OpenAI demanding to know whether their songs are being used to develop artificial intelligence (AI) systems.
Sony Music, which represents artists such as Beyoncé and Adele, prohibits the use of its songs to train, develop or monetize AI without permission.
In a letter to more than 700 companies, Sony said it “has reason to believe” the recipients “may already be misusing” its music.
The BBC has contacted Google, Microsoft and OpenAI for comment.
Sony has given companies a deadline to respond and said it will enforce copyright “to the fullest extent permitted by applicable law,” including the EU's upcoming AI law.
AI trained
Sony's complaint is part of a larger debate that has intensified since the advent of the latest generation of AI-powered tools. That is, what data was it trained on and did you ask for permission to use it?
For example, a chatbot may have “learned” to respond in a particular style by being shown billions of books, while an image-generating tool may have millions of existing images. Could have been given.
Similar software exists for creating music. Sony's complaint is that it can only be created after being exposed to millions of songs, some of which it owns the copyright to.
Google and OpenAI are both developing AI tools that can generate music, but it's unclear what data they used for training.
Copy not allowed
In a multi-page request, Sony asks each of hundreds of recipients to send the following:
Details of your own songs used to train the AI system
How to access songs – e.g. via online streaming services
How many copies of the song there are, whether copies still exist, and how long they have existed
Why did such a copy need to exist, if at all?
In the letter, the company said the copy could consist of only a portion of the song and that it was open to agreeing to a license agreement for future use.
However, it is unclear whether any of the AI companies contacted trained their technology using music owned by Sony, and if they did, it is unclear whether they actually violated any laws. .
Questions have arisen in both the EU and the US about whether training AI tools with such content constitutes copyright infringement, or whether it falls under fair use or “temporary copying” exceptions.
Nana Nwachukwu, a lawyer at AI ethics firm Saidot, said that under current EU rules, using copyrighted music to train an AI model “may constitute copyright infringement.” Ta.
He said “exceptions exist” for companies that can “legally access” music only if it's in the public domain and not behind a paywall, or if it's licensed for AI training. .
However, upcoming EU AI legislation will change this.
“This legislation requires that all general-purpose AI models… comply with strict documentation and transparency requirements,” she said.
“This includes obligations to publicly disclose detailed summaries of the training data used, to comply with copyright holder opt-outs, and to ensure that all data use complies with EU copyright law. Includes obligations.”
The issue has been heard in U.S. courts, with several lawsuits individually representing people including Game of Thrones author George R.R. Martin, comedian Sarah Silverman, and The New York Times. The case will be heard in court.
Meanwhile, as far as music is concerned, Sony's rival Universal Music has filed a lawsuit against Antrophic in the US, accusing the AI company of infringing copyright over the lyrics of hundreds of songs.
